Practical mobile sensing applications with privacy protection

Abstract

Mobile sensing is becoming an important part of people’s daily life. Thanks to various sensors equipped on devices such as smartphones and wearable devices, a wide range of human behaviour and context information can be obtained in real time. These information can be further used to support a large number of mobile applications which bring great convenience to people’s life, such as vehicle navigation, location-based advertising and taxi-hailing. However, some topics remain hard to be solved. In this thesis, we first propose a mobile sensing application – urban taxi-sharing. Then we address two challenging problems for practical mobile applications – privacy leakage and cold-start problem. In the first work, we design a QoS-aware taxi-sharing system named QA-Share. QA-Share allows occupied taxis to pick up new passengers on the fly, promising to reduce waiting time for taxi riders and increase productivity for drivers. Taxi-sharing can also bring significant social and environmental benefit, such as relieving traffic jams and saving energy consumption. We address two important challenges. First, QA-Share aims to maximise driver profit and user experience at the same tim. Second, QA-Share continuously optimises these two metrics by dynamically adapting its schedule as new requests arrive, without entering an oscillation state. Most mobile sensing applications rely on the sensing information shared by the crowd. Meanwhile, it has raised concerns for location privacy. Users may wish to prevent privacy leak and publish as many non-sensitive contexts as possible for better user experience of the application. Simply suppressing sensitive contexts is vulnerable to the adversaries exploiting spatio-temporal correlations in users’ behaviour. In the second work, we present a novel data sharing scheme PLP, which preserves privacy while maximizes the amount of data collection by filtering a user’s context stream. The experimental results show that PLP efficiently protects privacy without sacrificing much utility. Many context sensing methods are achieved by learning the relationship between context information and original sensing data of smartphones. However, manually labelling samples collected from smartphone users is time-consuming, labor-intensive and money-consuming. It’s hard to collect enough labeled samples at first. This is also known as the cold-start problem. In the third work, we propose iSelf, which can predict context information in cold-start conditions with smartphones. We take emotion detection as an example and show that our method can achieve high accuracy with only a few labeled samples.

Date of Award	2015
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology